|
I could use some of your feedback for a thing before I put more work into this. Want it to actually be useful and entertaining to folks.
|
# ? Jan 8, 2016 14:32 |
|
|
# ? May 16, 2024 19:05 |
|
Forever_Peace posted:I could use some of your feedback for a thing before I put more work into this. Want it to actually be useful and entertaining to folks. Minor thing: on the last graphic I would cut the x<0 left side off, since it's sort of artificial (if an RB actually had the ball 10 yards behind the line of scrimmage I wouldn't expect them to get another 13). One statistic that would be interesting to me is: just based on the distribution for an average back (or just with the overall distribution / i.e. a slightly above average back), if a team started on their own 20 and called running plays for every down and didn't punt or kick, how likely would they be to score a TD? Obviously not accounting for how predictable and ineffective this would probably be in practice, I just think it would be interesting. Also I think you laid the groundwork for the passing version follow-up pretty well, where you could just do the same graphics with less text etc. Would like to see the same thing I described in the last paragraph for that, too, though I guess you would have to factor in turnovers.
|
# ? Jan 8, 2016 15:39 |
|
Forever_Peace posted:I could use some of your feedback for a thing before I put more work into this. Want it to actually be useful and entertaining to folks. It looks solid. One thing I would say is if you're going to get in to predictive modeling with this I would use time series modeling instead of standard GLM stuff, especially ARCH models. I wouldn't be surprised if the expected YPC and variance of YPC are materially correlated for a given RB.
|
# ? Jan 8, 2016 16:14 |
|
pangstrom posted:I think it's really good. Not really a critique, but I think it's almost a better way to teach a football fan about descriptive statistics/distributions/graphics than it is to teach a person about running play expectations. Thanks! You think the Ernie thing works? And it's funny, I'm starting to really wish I had done some of thus stuff in time to use it in my stats class. Particularly the chapter on sampling and resampling - I sort of accidentally created a beautiful illustration of the central limit theorem. I'm a little worried I'm not striking the right balance between explanation and information, though. I think a lot of the folks that are most interested in this don't really need the explanation, but I went explanation-heavy anyways just to make it more accessible. Did I overshoot? Less explanation would certainly make these shorter... Reik posted:It looks solid. One thing I would say is if you're going to get in to predictive modeling with this I would use time series modeling instead of standard GLM stuff, especially ARCH models. I wouldn't be surprised if the expected YPC and variance of YPC are materially correlated for a given RB. I'd actually love some input on the modeling. So far, I'm using an ensemble of a GAM and a random forest model. The GAM: gam(log_yds~homefield+defense*year+s(yards_to_go,k=10,by=down)+s(yards_til_goal,k=50)+s(score_dif,k=10, by=quarter)+s(secondsleft, k=50,by=winning),data=df_rb_3shift) The random forest: randomForest(log_yds~year+yards_to_go+down+yards_til_goal+homefield+quarter+winning+score_dif+defense,data=df_rb_3shift) Both use the same outcome measure of "log_yds", or the log of the yardage that has been right-shifted by three yards (the "3shift" data set). Essentially, "0" is three yards behind the line of scrimmage (where running backs generally get/have the ball and are ready to hit a hole), and the yardage past that point is log-transformed (to make it normally distributed, and to de-emphasize yardage differences for long runs). Neither captures the dependencies across offenses or individual running backs, because I wanted to compare marginal yardage of particular backs after the model-fitting. But perhaps a fixed factor for player would be the way to go here? That would technically also be interpreted as marginal to the situational factors. I'm not really sure. I'm not too familiar with ARCH models and time-series analysis (though I've done a few ARIMA models). Is the suggestion here to actually track these factors literally over time for each player? Like, week by week?
|
# ? Jan 8, 2016 16:36 |
|
This is probably not the thread for this, but is there some kind of random-effect version of ARMA/VAR/GARCH/whatever models? Forever_peace, it's awesome. There's no perfect amount of explanation that works best for everyone, so just keep going as you are! That would be my advice at least.
|
# ? Jan 8, 2016 16:43 |
|
I do think the Ernie thing works, it took me a bit because I was expecting it to be more of a "conversation" between the author and him, whereas he's doing more color commentating/chipping in, but I don't think that's a problem and it's probably better the way you have it. I'm a bad writer so my opinion there isn't worth much.
|
# ? Jan 8, 2016 16:47 |
|
Forever_Peace posted:I'd actually love some input on the modeling. Toss your codes up on github and you can probably get a lot of input from here and the various programming/stats threads.
|
# ? Jan 8, 2016 16:55 |
|
Forever_Peace posted:I'm not too familiar with ARCH models and time-series analysis (though I've done a few ARIMA models). Is the suggestion here to actually track these factors literally over time for each player? Like, week by week? My thought is there are situations that arise that can have a material effect on performance, maybe the offense is in a good rhythm and they're doing well or maybe their defense has been on the field for 40 minutes already and they're worn down. Using each run as a data point you could pick up on these trends using previous runs in that game as inputs to the next run. You would also use runs from previous games as inputs, but they probably wouldn't have as much impact. Once you've projected runs on a per-run basis you can total them up by week. GARCH models are used to get around the assumption that your variables have a constant variance. When they don't, it's called heteroskedasticity and it throws regular ARIMA models for a loop. Essentially a GARCH model predicts variance based on based variance then uses the variance as an input in the expected value prediction. It's very popular in finance since a 5% return on a risk-free asset is valued more than a 5% expected return on a risky asset, so you need some way to differentiate the 5% expected returns.
|
# ? Jan 8, 2016 17:36 |
|
just submitted my lineup for the yahoo champion of champions thing this is the year i finally repay my student loans
|
# ? Jan 8, 2016 20:12 |
|
whatis posted:just submitted my lineup for the yahoo champion of champions thing
|
# ? Jan 8, 2016 20:33 |
|
Forever_Peace posted:I could use some of your feedback for a thing before I put more work into this. Want it to actually be useful and entertaining to folks. Just wanted to say I enjoyed it and I think the Ernie thing works. If I had to come up with a criticism for it, I'd say reiterate some of the important takeaways in the chapter as part of your conclusion. The rules of run quantities for example is a great tidbit and summary of some stats that you explore in the post, but it's in the middle of the post itself and by the time I've finished the chapter, I've already forgotten it. You start the chapter off around the question "What should I expect from a run play?" and get into percentiles with some key tips (rules of run quantities) and then into yards left (key breakaway point at 5 yards being one good takeaway), but a summary in the conclusions that just quickly reiterates some of the key points from those would tie it all up nicely, I think. I guess my point is you don't have too many words, and the words you have are good and chunked well enough to make the information easily legible and digestible. Keep the theme or main question you introduce in the beginning relevant throughout the chapter, try to build up some key takeaways throughout the chapter, then at the end reiterate those takeaways so I remember them once I close your post and go beg for trades in my dynasty leagues.
|
# ? Jan 8, 2016 21:56 |
|
Thanks folks. I really like the idea of a chapter summary at the end. I'll definitely start doing that. Github is also probably a good idea. I'll look into making a new account that isn't tied to my real name.
|
# ? Jan 9, 2016 16:23 |
|
Hazo posted:Same here. Put barely any thought into it but it's free so why not. Yep. Zero thought into it, but boy wouldn't it be nice to win.
|
# ? Jan 9, 2016 21:43 |
|
cool kids inc. posted:Yep. Zero thought into it, but boy wouldn't it be nice to win. I gotta wedding to pay for lets do this thing. I also think you should get more than one entry if you win more than one league.
|
# ? Jan 9, 2016 22:07 |
|
Also in the Yahoo: Champion of Champions, it is my first time playing a daily. Since there are like 270k people, I view it as finding a free lotto ticket on the ground. Odds are it is worthless, but might as well fill it out. So far cannot complain with 41.7 from Ware and KC. Pain of Mind fucked around with this message at 01:39 on Jan 10, 2016 |
# ? Jan 10, 2016 01:34 |
|
Anybody make it in the Yahoo champions thing? I barely even finished in the top half: Russel/Ware/Gio/AB84/MarvinJones/Cobb/Reed/DJax/SEA Of these, only Ware/AB84/Cobb/Reed made regular appearances in the top 10. Get this: the winnings teams mostly played two TEs (Reed and Kelce). That hadn't even occurred to me.
|
# ? Jan 11, 2016 16:07 |
|
I got like 263,000/266,000, oops. Although I did get my first 4-figure score on DraftKings! Yes, from a Reed/Kelce/Chiefs lineup.
|
# ? Jan 11, 2016 16:11 |
|
79582nd apparently, which is decidedly not good, but not loving terrible either. I don't move on to the semifinals. Put in Peterson hoping something would happen and it didn't, and Roethlisberger got his poo poo rocked and was out for a sizable chunk of game. Brown, Reed, and KC defense were the only players of mine to do much of note. Ah, well.
|
# ? Jan 11, 2016 16:17 |
|
I was 55,000/267,000 going into the late game, and ended up below 185,000th. Roethlisberger/Jeremy Hill/Christine Michael/Doug Baldwin/Tyler Lockett/James Jones/Tyler Eifert/Charmander/SEA. Oh well! The winning team, if anybody was wondering, was: Cousins/Alfred Blue/Fitzgerald Toussaint/Martavis Bryant/AJ Green/Randall Cobb/Jordan Reed/Travis Kelce/KC
|
# ? Jan 11, 2016 16:19 |
|
I made it, I got 9153. 125.46 points with: Cousins Michael Peterson Brown Garcon Wheaton Ware Reed KC I thought I was dead due to Peterson and Wheaton getting a combined 6 points, but there were so many disappointing players this week than having 2 did not hurt too bad, especially with best in position at KC, Cousins, and Reed. I checked what the cutoff was for making the next round, and it was 118.56. There were also ~2500 teams within 1 point of the cutoff. Is there any way to look up the % owned for a player in advancing teams? I want to see how high KC is.
|
# ? Jan 11, 2016 17:27 |
|
I think they've already taken down the teams if you didn't make the cut, but I remember seeing percent owned next to the player names just by clicking on the team.
|
# ? Jan 11, 2016 21:27 |
|
Hazo posted:I think they've already taken down the teams if you didn't make the cut, but I remember seeing percent owned next to the player names just by clicking on the team. Yea, I saw that, I meant % owned for teams that advanced to the next round, not total.
|
# ? Jan 11, 2016 21:43 |
|
Yeah, I didn't make it very far. Till next year. On to Fantasy Nascar
|
# ? Jan 12, 2016 01:37 |
|
Pain of Mind posted:I made it, I got 9153. 125.46 points with: Yeah, it was a decidedly bad week for scoring in general. I did pretty well, but I missed the cut by about 18 points without having KC on defense, even though I had spare cash to pick them up. Oh well
|
# ? Jan 12, 2016 17:56 |
|
Did pretty well with this lineup, hope we move on again this week! E: Flex Blue Chiefs D
|
# ? Jan 12, 2016 18:22 |
|
I was browsing random teams at the bottom figuring most would be inactive, but a lot actually had a full roster. I feel like a lot of people read a DFS strategy article and took the "try and get low % owned players" but failed to apply some sort of sanity check to it. Weeden and a bunch of TE2s and WR4s didn't win, shocker.
|
# ? Jan 12, 2016 19:18 |
|
Pain of Mind posted:I was browsing random teams at the bottom figuring most would be inactive, but a lot actually had a full roster. I feel like a lot of people read a DFS strategy article and took the "try and get low % owned players" but failed to apply some sort of sanity check to it. Weeden and a bunch of TE2s and WR4s didn't win, shocker. Gio and Jamison Crowder plz. And Nick Novak
|
# ? Jan 12, 2016 19:20 |
|
Spoeank posted:Gio and Jamison Crowder plz. The first iteration of my lineup and Gio and Crowder also, not sure it would have changed much since I replaced Crowder with Wheaton who probably did worse. Also, I put together my team for week 2 of yahoo's champions of champions. I feel like half of my team is going to be owned by 60% of people though, some players seemed to have much more value/$ than others.
|
# ? Jan 12, 2016 19:54 |
|
Charity league raised nearly $4500 to send to SPLC. A Good Season.
|
# ? Jan 14, 2016 13:38 |
|
Paging Dandy Kaiser http://www.fantasypros.com/2016/01/fantasy-football-lessons-learned-wrte/ (the last one)
|
# ? Jan 15, 2016 20:22 |
|
quote:Going into the season, the tight end position was framed as Gronk, Graham/Kelce/Olsen, then Bennett/Witten, then everyone else. It did not play out this way. Only three of the top 10 tight ends (by fantasy points per game) were drafted in the top 100 picks. The position has grown very fungible, and unless you get Rob Gronkowski, it isn’t worth drafting a tight end until at least the ninth round. Even then, you may be overpaying for Gronk. quote:Going into the season, the tight end position was framed as Gronk, Graham/Kelce/Olsen, then Bennett/Witten, then everyone else. It did not play out this way. Only three of the top 10 tight ends (by fantasy points per game) were drafted in the top 100 picks. The position has grown very fungible, and unless you get Rob Gronkowski, it isn’t worth drafting a tight end until at least the ninth round. Even then, you may be overpaying for Gronk. quote:Going into the season, the tight end position was framed as Gronk, Graham/Kelce/Olsen, then Bennett/Witten, then everyone else. It did not play out this way. Only three of the top 10 tight ends (by fantasy points per game) were drafted in the top 100 picks. The position has grown very fungible, and unless you get Rob Gronkowski, it isn’t worth drafting a tight end until at least the ninth round. Even then, you may be overpaying for Gronk. quote:Going into the season, the tight end position was framed as Gronk, Graham/Kelce/Olsen, then Bennett/Witten, then everyone else. It did not play out this way. Only three of the top 10 tight ends (by fantasy points per game) were drafted in the top 100 picks. The position has grown very fungible, and unless you get Rob Gronkowski, it isn’t worth drafting a tight end until at least the ninth round. Even then, you may be overpaying for Gronk. quote:Going into the season, the tight end position was framed as Gronk, Graham/Kelce/Olsen, then Bennett/Witten, then everyone else. It did not play out this way. Only three of the top 10 tight ends (by fantasy points per game) were drafted in the top 100 picks. The position has grown very fungible, and unless you get Rob Gronkowski, it isn’t worth drafting a tight end until at least the ninth round. Even then, you may be overpaying for Gronk. e:
|
# ? Jan 15, 2016 20:25 |
|
wait i'm not done yet me atm irl
|
# ? Jan 15, 2016 20:30 |
|
I drafted Gronk in the second round and won my league. He was also the second highest scorer on my team.
|
# ? Jan 15, 2016 20:36 |
|
Spoiler alert about my link re: Dandy Kaiser that was me, I wrote that
|
# ? Jan 15, 2016 20:37 |
|
Eifert and Delanie Walker were real cool to have off waivers, too bad the rest of my teams sucked balls
|
# ? Jan 15, 2016 20:39 |
|
Spoeank posted:Spoiler alert about my link re: Dandy Kaiser
|
# ? Jan 15, 2016 20:43 |
|
SurgicalOntologist, pangstrom, and pmchem check your PMs. I am undergoing babbys first github shiny app distribution and could use an assist.
|
# ? Jan 16, 2016 05:52 |
|
I drafted Gronk and traded him a month before playoffs. I played the guy who had him against me in the championship. My waiver wire pickup (that week) and homer pick, Zachary Ertz, killed Gronk. I win.
|
# ? Jan 16, 2016 09:45 |
|
I know some of you guys moved on in the Champion of Champions DFS. Good luck goons, here's my hopeful entry:
|
# ? Jan 16, 2016 23:06 |
|
|
# ? May 16, 2024 19:05 |
|
Rick Grimes posted:I know some of you guys moved on in the Champion of Champions DFS. Good luck goons, here's my hopeful entry: I started the wrong white guys.
|
# ? Jan 17, 2016 07:19 |